Problem
Research questions and friction points this paper is trying to address.
state encodings
collective dynamics
synchronization
language-model agents
effective interaction law
Same physical state, different collective dynamics: state encodings select synchronization outcomes in language-model agents
🤖 AI Summary
This study demonstrates that the collective synchronization behavior of language model agents depends critically on the choice of state encoding rather than being solely determined by the underlying physical system. By systematically varying only the state representation—such as low-order circular moments versus histograms—in a fixed ring-based synchronization task and analyzing behavioral replay trajectories, the authors reveal that the encoding scheme effectively functions as a model-specific interaction rule that profoundly shapes emergent collective dynamics. Experiments show that GPT achieves perfect (100%) synchronization with circular moment encoding but fails entirely under histogram encoding, whereas Claude exhibits the opposite pattern. These findings establish, for the first time, that state encodings are not neutral interfaces but instead serve as pivotal mechanisms that directly influence agents’ decision distributions and collective outcomes.
📝 Abstract
Language-model agents act on state encodings of their environment, yet these are treated as interchangeable interfaces. Using pretrained language models, we designed a circular-synchronization experiment applying a state-encoding intervention while holding the physical system fixed: each agent sees only a summary of its neighbours' relative phases and chooses to advance, stay or retard. Encoding that state as low-order circular moments rather than as a histogram selected different collective outcomes. In GPT the moment encoding synchronized the population in 6/6 seeds and the histogram encodings in 0/6; the effect replicated in Claude but reversed direction. Replaying identical fields shifted each agent's advance/stay/retard probabilities far beyond within-encoding repeat variation, in GPT, Claude and Gemini; in GPT, presentation alone shifted the operator with the moment values fixed. State encodings therefore form part of a model-dependent effective interaction law, not a neutral interface.
Innovation
Methods, ideas, or system contributions that make the work stand out.
state encoding
collective dynamics
synchronization
language-model agents
circular moments
🔎 Similar Papers
No similar papers found.
